dagster: Command for Claude Code

Migrate NamedTuple to @record

Convert the specified NamedTuple class to use @record from dagster_shared.record.

Use @record (simple case)

When __new__ only contains check.*_param calls with no coercion:

@record
class Foo:
    x: str
    y: Optional[int] = None

Use @record_custom when:

1. Any custom logic - validation, invariants, conditionals
2. Collection coercion - check.opt_*_param coerces None → empty collection

Coercion functions requiring @record_custom:

• check.opt_mapping_param / check.opt_dict_param → {}
• check.opt_list_param / check.opt_sequence_param → []
• check.opt_set_param → set()

Callers may pass None explicitly even without a default - preserve the coercion:

@record_custom
class Foo(IHaveNew):
    tags: Mapping[str, str]

    def __new__(cls, tags: Optional[Mapping[str, str]] = None):
        return super().__new__(cls, tags={} if tags is None else tags)

When using @record_custom, there should be no defaults set on the field definitions, only the __new__ arguments.

Conversion steps

1. Remove NamedTuple inheritance and inline field definitions
2. Add @record or @record_custom decorator
3. Move fields to class body with type annotations
4. Delete check.*_param calls (auto-generated) unless coercion needed
5. Keep only custom validation in __new__ for record_custom
6. Handle NamedTuple method usage - search codebase for calls to ._replace(, ._asdict(, ._fields on this class:
   • Option A: Update callsites to use @record equivalents (from dagster_shared.record):
     • obj._replace(...) → copy(obj, ...) or replace(obj, ...)
     • obj._asdict() → as_dict(obj)
     • Cls._fields → get_record_annotations(Cls).keys()
   • Option B: Add LegacyNamedTupleMixin to preserve NamedTuple API compatibility
7. Use kwargs only in super().__new__() for record_custom
8. Run just ruff

Positional args

For public APIs needing positional arg support: @record(kw_only=False)